In this article
A new dish that sells well feels like proof the menu is working. Most of the time it is proof of nothing at all — because a POS report shows units sold, and it has no way of showing units taken from somewhere else.
Somewhere on a menu right now, a new dish is quietly moving 45 covers a week and being treated as a hit. Nobody has checked what happened to the dish that used to sit next to it on the card, because nothing in a till report prompts that question. Sales is the only number the system counts.
This is the same problem retail and consumer-goods companies have been measuring for decades under a plain name: cannibalization. Launch a new product and some share of its sales is genuinely new demand, and some share is a customer who was always going to buy something from you, buying this instead of that. Nobody in hospitality writes it down, because a kitchen counts plates and a POS counts orders — neither one asks where an order would otherwise have gone.
It matters more on a menu than almost anywhere else, because a menu is a fixed number of decisions a night. A guest orders one main course, not one-point-two. A dish that finds 45 buyers a week did not create 45 new decisions; at most it redirected the decisions that were already being made, at a different margin than before.
This guide walks through the four numbers that separate a genuine win from a dish that only looks like one: how much of its sales are diverted, how many covers actually grew, what that swap did to your profit in money, and the point at which a dish stops paying for its own place on the card. At the bottom you put your own new dish and up to three neighbours through the same arithmetic.
Why a busy new dish can still be a quiet loss
A menu does not add capacity when you add a line to it. The room seats the same number of people, the kitchen has the same number of covers to send out in a service, and every guest still orders roughly one main course. So when a new dish sells, the first and most useful question is not "did it sell" — it is "instead of what".
Nothing in ordinary reporting answers that question, because nothing links two menu items together. Your POS knows the mushroom risotto sold 45 times last week. It has no field for "and the tagliatelle two rows up sold 24 fewer times than it did the month before you launched it" — that comparison exists only if somebody thinks to run it, dish by dish, by hand.
And even once you notice the drop, the arithmetic is not "subtract one from the other". A diverted cover is not lost — the guest still ordered a main course, just a different one — so what actually changed is the DIFFERENCE between what the new dish earns on that cover and what the old one used to earn on it. A new dish with a thinner margin than what it replaced can be selling well and making you poorer at the same time, and a spreadsheet that only tracks units sold will never show you that.
The ultimate guide Menu Engineering & Drinks: 6 Steps to More Margin Stars, Plowhorses, Puzzles and Dogs — and everything else your menu can earn you. Open the guideThe 4 numbers that tell you what really happened
They build on each other in order. The first tells you how much of the new dish's success is borrowed. The second tells you what actually grew. The third turns both into money. The fourth tells you the ceiling — the diversion rate above which the dish stops paying for its own slot on the card, whatever its covers say.
1. Diversion rate — how much of this dish's success is borrowed
Take the 55-seat brasserie this guide uses throughout. A wild mushroom risotto launches and settles at 45 covers a week — a clean, comfortable number for a new main. Look at what sits near it on the card and the picture changes: the truffle tagliatelle drops from 38 covers a week to 14, and the plainer mushroom risotto it was quietly replacing drops from 22 to 6. A pan-seared cod two lines down barely moves, 18 to 17 — not everything near a launch is a victim, which is exactly why you check dish by dish rather than assuming.
Add up what dropped: 24 plus 16 plus 1 is 41 covers a week that vanished from the dishes around the new one. Divide that by the new dish's own 45 covers and you get a diversion rate of 91%. Nine out of every ten covers the new risotto "sold" were already being sold by something else on the menu — they simply moved.
One guard belongs in this number: a neighbouring dish can drop for reasons that have nothing to do with your launch — a bad review, a supplier problem, a different season. So the arithmetic never attributes more drop to the new dish than the new dish itself sold: if every neighbour's drop adds up to more covers than the new dish generated, every neighbour's share is scaled down by the same factor rather than one being blamed for all of it.
Weekly covers before and after launch. The new dish's own 45 covers, set against three dishes that already sat on the card.
Two neighbours dropped hard, a third barely moved. That is normal — a launch rarely diverts evenly, and checking dish by dish is the only way to see which one actually paid for the new arrival.
2. True incremental covers — what actually grew
Once you know the diversion rate, the covers that are genuinely new fall straight out of it: 45 total minus the 41 diverted is 4 covers a week of real growth. Not 45. Four.
This is the number a POS report will never show you, because a POS report has no concept of "instead of". It is also the number that should set your expectations before you launch: a new main that only ever finds four truly new decisions a week was never going to transform your covers count, however well it "sells" once the diverted volume is counted in.
It is not a verdict on its own, though — four incremental covers on a thin margin is nothing, and four incremental covers on a strong one can matter. What decides that is the next number.
3. Net margin swing — what the swap actually did to your profit
A diverted cover is not a loss, because the guest still ordered a main course — they ordered this one instead of that one. So the effect on your profit is the DIFFERENCE between what the new dish earns on that cover and what the dish it replaced used to earn, not the old dish's whole margin counted as vanished.
Run the full 55-seat brasserie through it. The new risotto is on the card at €19.45 including VAT, costs €6.10 to plate, and at 13.5% catering VAT that leaves a margin of roughly €11.04 per cover. The truffle tagliatelle it displaced kept about €12.61 a cover; the plainer risotto about €10.50. Weight those by how many covers each one actually lost, add the four genuinely new covers at the new dish's own margin, and the whole picture nets out to about €16 a week — roughly €848 a year.
Put that beside what the dish looks like on paper: 45 covers a week at €19.45 is €875 of gross weekly revenue including VAT. The genuine, cannibalization-adjusted gain is worth about 1.9% of that. This dish is not the disaster a raw margin comparison might suggest, and it is nowhere near the hit its own sales figure implies either — it is a narrow, real win, sitting much closer to break-even than its covers count would ever tell you.
€19.45 including VAT, split into VAT, the plate's own cost, the margin it took from its neighbours, and what is genuinely left over.
The red band is not a cost you paid — it is margin that used to belong to another dish and now belongs to this one instead. It still counts against the new dish, because that margin no longer arrives twice.
4. Break-even diversion rate — the ceiling this dish is testing
The three numbers above describe what happened. This one tells you how much room is left before it stops being worth it — and, unlike the others, it does not depend on how many covers the dish actually sells. Solve the same arithmetic backwards for the diversion rate at which the swap nets to exactly zero, and the volume cancels out of the equation entirely: it is simply the new dish's own margin divided by the average margin of what it is displacing.
For the brasserie's risotto that works out to about 94%. Its actual diversion rate is 91% — inside the ceiling, but by only about three percentage points. A slightly cheaper cut of mushroom, a few cents more on the price, or one more guest a week choosing the tagliatelle again would be enough to tip this exact dish from a narrow win into a net loss, however healthy its 45-cover sales figure keeps looking.
That is the real use of this number: it is not a verdict on one dish, it is a ceiling you can check any future one against before it earns a permanent slot on the card. A dish whose own margin comfortably beats what it is likely to displace can survive heavy diversion. A dish whose margin is weaker than its neighbours cannot survive much diversion at all, no matter how well it appears to sell.
Run your own new dish through it
Fill in your new dish and up to three dishes that sit near it on the card — their price, their cost, and their weekly covers before and after the launch. The five entries are pre-filled with the brasserie's own figures above so you can see immediately how they read; overwrite them with yours.
You get the same four numbers worked out for your own menu: the diversion rate, the covers that genuinely grew, what the swap is worth in money, and the ceiling your dish is testing against.
Menu cannibalization scan
One new dish, up to three neighbours, and the four numbers behind them.
—
The calculation assumes every neighbour's drop that exceeds the new dish's own covers is due to something else, and scales it down accordingly. Everything is worked out in your browser; nothing is sent or stored.
Two things worth keeping in mind when you read your own numbers. A diverted cover is never counted as a full loss of the old dish's margin — it is counted as the DIFFERENCE between the two margins, because the guest was always going to order a main course. And the break-even ceiling does not depend on how many covers you sell: a dish with a strong enough margin can survive very heavy diversion, and a dish with a weak one cannot survive much at all.
None of this argues against ever launching a dish that overlaps with an existing one. A menu that never changes goes stale, and some diversion is the price of keeping it interesting. What it argues for is checking — because the difference between a dish worth keeping and one worth dropping is very rarely visible in its own sales figure.
What you do with it this week, this month and this quarter
You cannot run this check on every dish at once, and you do not need to. This order finds the ones that matter first.
This week — find the pairs worth checking
- List every dish launched in the last six months, and for each one, name the one or two dishes that sit nearest to it on the card by ingredient or by course.
- Pull weekly covers for each pair from four weeks before the launch and the four most recent weeks.
- Put your newest, biggest-selling dish through the scan above first — it is the one most likely to be hiding the largest swing.
- Note which neighbours barely moved. Those pairs are not cannibalization and do not need a second look.
This month — turn the swing into a decision
- For every dish whose diversion rate sits within ten points of its break-even ceiling, decide now rather than later: adjust the price, tighten the plate cost, or accept it as a narrow win.
- For any dish already a net loss, change one variable — usually cost or price, rarely the dish itself — and re-run the scan before touching the card again.
- Where a new dish is a genuine win, ask what made it one: a real margin advantage, or a genuinely new audience it reached that the neighbours never did. That is worth repeating on purpose.
- Retire the dish it displaced only once you are certain the swap nets you more, not just sells more.
This quarter — build the check into the launch itself
- Before your next new dish goes live, name its likely neighbours in advance and note their current weekly covers — the comparison is far cheaper to run when you already have a starting point.
- Put the scan on your calendar six to eight weeks after every future launch, before the new dish's numbers are treated as settled.
- Read this alongside your menu engineering matrix rather than instead of it: a dish can score well as a Star on its own popularity and margin and still be quietly displacing a higher one.
- And once a dish's covers are genuinely incremental, cost it properly with the recipe costing tool so its margin is not a guess the next time you run this check.
A dish that sells is not the same as a dish that pays
The brasserie's new risotto sells 45 covers a week and would pass almost any glance at a sales report. Run the same numbers through this arithmetic and it is a dish clearing about €848 a year, on a ceiling it is already within three points of — a genuine win, and a far smaller one than its own popularity suggests.
That gap between what a dish appears to do and what it actually does is not an argument against trying new things on the menu. It is an argument for checking the dish next to it before deciding the new one is the hit its covers make it look like.
Do the same with the rest of what is already on the card. Menu engineering tells you which existing dishes earn their place using their own popularity and margin; this tells you what a new one did to the dishes it landed beside. Between the two, you are reading the whole card, not just the newest line on it.